Minimax Image UnderstandingSAFE

High-performance AI coding agent CLI written in Rust with zero unsafe code

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: minimax-image-understanding
description: Analyze images using AI with the understand_image tool
---

# MiniMax Image Understanding Skill

Use this skill when you need to analyze, describe, or extract information from images.

## How to Use

Call the `understand_image` tool directly with a prompt and image URL:

```
understand_image({
  prompt: "Your question about the image",
  image_url: "https://example.com/image.png"
})
```

## When to Use

Use `understand_image` when:

- **Screenshots**: Error messages, UI issues, code in screenshots
- **Visual content**: Photos, diagrams, charts, graphs
- **Documents**: Extracting text from images (OCR), understanding layouts
- **UI/UX analysis**: Evaluating designs, identifying components
- **Visual debugging**: Understanding visual bugs or layout issues

## When NOT to Use

Do NOT use `understand_image` when:

- **Image is already described** in the conversation
- **The image is a simple icon** or emoji you recognize
- **No image is provided** or the image URL is inaccessible
- **Redundant with existing context** (e.g., file contents already visible)

## Usage

```
understand_image({
  prompt: "What do you see in this image?",
  image_url: "https://example.com/screenshot.png"
})
```

## API Details

**Endpoint**: `POST {api_host}/v1/coding_plan/vlm`

**Request Body**:
```json
{
  "prompt": "Your question about the image",
  "image_url": "data:image/jpeg;base64,/9j/4AAQ..."
}
```

**Response Format**:
```json
{
  "content": "AI analysis of the image...",
  "base_resp": {
    "status_code": 0,
    "status_msg": "success"
  }
}
```

## Image Processing

The tool automatically handles three types of image inputs:

1. **HTTP/HTTPS URLs**: Downloads the image and converts to base64
   - Example: `https://example.com/image.jpg`

2. **Local file paths**: Reads local files and converts to base64
   - Absolute: `/Users/username/Documents/image.png`
   - Relative: `images/photo.png`
   - Removes `@` prefix if present

3. **Base64 data URLs**: Passes through existing base64 data
   - Example: `data:image/png;base64,iVBORw0KGgo...`

## Image Formats

Supported:
- **JPEG** (.jpg, .jpeg)
- **PNG** (.png)
- **WebP** (.webp)

Not supported:
- PDF, GIF, PSD, SVG, and other formats

## Crafting Effective Prompts

### For Descriptions
- "Describe what's in this image in detail"
- "What is the main subject of this image?"
- "Describe the visual style and composition"

### For Code/Technical
- "What code is shown in this screenshot?"
- "Extract all text from this image"
- "Identify the UI framework/components used"

### For Analysis
- "Analyze this UI design. What is working well and what could be improved?"
- "What emotions or mood does this image convey?"
- "Compare this design to Material Design principles"

### For OCR/Text Extraction
- "Extract all text from this image"
- "Read the error message in this screenshot"
- "What does the label say in this image?"

## Examples

### Error Analysis
```
understand_image({
  prompt: "What is the error message and where is it located in this screenshot?",
  image_url: "./error-screenshot.png"
})
```

### Code Screenshot
```
understand_image({
  prompt: "What code is shown in this screenshot? Please transcribe it exactly.",
  image_url: "https://example.com/code.png"
})
```

### Design Review
```
understand_image({
  prompt: "Analyze this UI design. What is working well and what could be improved?",
  image_url: "https://example.com/mockup.png"
})
```

### OCR
```
understand_image({
  prompt: "Extract all text from this image",
  image_url: "/Users/username/Documents/scan.png"
})
```

## Tips

1. **Be specific** in your prompt about what you want to know
2. **Mention format** if you need structured output (e.g., "list all elements")
3. **Include context** if the image is part of a larger task
4. **For screenshots**, specify if you need full-page or just a specific area
5. **Complex analysis** may trigger a confirmation prompt (analyze, extract, describe, recognize, transcribe, read)

## Error Handling

- **Status code 1004**: Authentication error - check API key and region
- **Status code 2038**: Real-name verification required
- **Invalid image**: File doesn't exist or URL is inaccessible
- **Unsupported format**: Image format not in JPEG, PNG, WebP

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (1)

.beads/beads.jsonl

Gates applied: no_behavioural_pass.